Home

Many studies exist on distance to health facilities, but in LMIC actual health data is not collected in a systematic fashion. Using the HDSS data and framework, we are able to have location-based health data which can help in decision-making and can enhance our understanding of the spatio-temporal dimensions of monitoring health and improving outcomes in post-war recovery regions. We used these data and several spatial analysis techniques to assess the degree to which health outcomes in Gulu (specifically, malnutrition) correlate with geographic accessibility. The chosen methods are all backed by literature.

Seasonality was identified as a potential confounding variable in our health outcome data. In Uganda, the 'hungry season' is a period from May to July that is characterized by depleted food resources and labour-intensive agricultural activity, resulting in widespread hunger (Maranz et al., 2004; Schramm et al., 2016). To account for this, health measurements obtained during the hungry season were excluded from all analyses.

Kernel density smoothing was applied to local health outcome data to examine the spatial pattern of malnutrition within the study area. Kernel density smoothing is a statistical procedure frequently used in epidemiological analysis that produces a continuous surface that estimates the density of an event (in our case, malnutrition) within a defined window.

We also employed Local Moran's I to analyze our health data for statistically significant hotspots of malnutrition. This is a common cluster detection technique that measures the level of spatial autocorrelation between locations in the study area, with consideration given to both the distance between locations and their attribute values (Huang et al., 2014; Lee & Li, 2016). By implementing Local Moran's I, we were able to determine the locations of significant clusters or hotspots of underweight and stunting incidence, as well as outliers.

In addition to underweight and stunting incidence, separate hotspot analyses were performed on z-scores for body mass index (BMI) and height-for-age (HFA) and weight-for-length (WFL) values, typical anthropometric indices of malnutrition (World Food Programme, 2005). Only those individuals with z-scores less than or equal to -2 were considered, as this represents the threshold value for moderate to severe undernutrition in children (World Health Organization, 1997).

To examine the effect of differential geographic access to health care, a network analysis was performed with respect to households and health centers. This allowed us to create catchment areas for each health center and to determine which homes fall within health center catchment areas and which do not. We used the field "time" to derive a catchment area and our speed will be assumed at a basis of a walking speed of approximately 4km/h. The results generated catchment areas for each centre type (Health Centres II, III, and IV, and hospital), as well as an overall map, with a predetermined maximum time for each catchment of 60 minutes walking one-way.

Following the completion of our analyses, we combined the generated outputs to make qualitative assessments of the relationship between malnutrition and the spatial organization of health care in our study area.